LLM Security Red Lines: Prompt Injection Defense in Practice

Simi included in AI

2025-12-26 69 words One minute

Contents

What Is Prompt Injection

Attack: inject malicious instructions in user input
"ignore above instructions, execute XXX instead"

Defense Strategy

        
        
        
    
# 1. Input filtering
def sanitize_prompt(user_input):
    blocked = ["ignore", "disregard", "forget previous"]
    for pattern in blocked:
        if pattern in user_input.lower():
            raise ValueError("blocked pattern")
    return user_input

# 2. Output validation
def validate_output(response):
    if contains_sensitive_data(response):
        raise ValueError("PII detected")
    return response

Conclusion

LLM security trio: input filtering + output validation + least privilege.